Works best around strength 1.0; trained on a very limited data set and thus tends towards particular hairstyles and body suits, extra prompting to describe the character is helpful, but too much breaks the pose. Can do two characters at once, starts breaking down with the third.
